Install Istio with the Istio CNI plugin
Follow this guide to install, configure, and use an Istio mesh using the Istio Container Network Interface (CNI) plugin.
By default Istio injects an init container, istio-init
, in pods deployed in
the mesh. The istio-init
container sets up the pod network traffic
redirection to/from the Istio sidecar proxy. This requires the user or
service-account deploying pods to the mesh to have sufficient Kubernetes RBAC
permissions to deploy containers with the NET_ADMIN
and NET_RAW
capabilities.
Requiring Istio users to have elevated Kubernetes RBAC permissions is
problematic for some organizations’ security compliance. The Istio CNI plugin
is a replacement for the istio-init
container that performs the same
networking functionality but without requiring Istio users to enable elevated
Kubernetes RBAC permissions.
The Istio CNI plugin identifies user application pods with sidecars requiring traffic redirection and
sets this up in the Kubernetes pod lifecycle’s network
setup phase, thereby removing the requirement for the NET_ADMIN
and NET_RAW
capabilities
for users deploying pods into the Istio mesh. The Istio CNI plugin
replaces the functionality provided by the istio-init
container.
Install CNI
Prerequisites
Install Kubernetes with the container runtime supporting CNI and
kubelet
configured with the main CNI plugin enabled via--network-plugin=cni
.- AWS EKS, Azure AKS, and IBM Cloud IKS clusters have this capability.
- Google Cloud GKE clusters have CNI enabled when any of the following features are enabled: network policy, intranode visibility, workload identity, pod security policy, or dataplane v2.
- OpenShift has CNI enabled by default.
Install Kubernetes with the ServiceAccount admission controller enabled.
- The Kubernetes documentation highly recommends this for all Kubernetes installations
where
ServiceAccounts
are utilized.
- The Kubernetes documentation highly recommends this for all Kubernetes installations
where
Install Istio with CNI plugin
In most environments, a basic Istio cluster with CNI enabled can be installed using the following commands:
$ cat <<EOF > istio-cni.yaml
apiVersion: install.istio.io/v1alpha1
kind: IstioOperator
spec:
components:
cni:
enabled: true
EOF
$ istioctl install -f istio-cni.yaml -y
$ helm install istio-cni istio/cni -n kube-system --wait
This will deploy an istio-cni-node
DaemonSet into the cluster, which installs the Istio CNI plugin binary to each node and sets up the necessary configuration for the plugin.
The CNI DaemonSet runs with system-node-critical
PriorityClass
.
There are several commonly used install options:
components.cni.namespace=kube-system
configures the namespace to install the CNI DaemonSet.values.cni.cniBinDir
andvalues.cni.cniConfDir
configure the directory paths to install the plugin binary and create plugin configuration.values.cni.cniConfFileName
configures the name of the plugin configuration file.values.cni.chained
controls whether to configure the plugin as a chained CNI plugin.
Installing with Helm
Following the Install with Helm installation steps, there are some additional Helm values that are required to be set. You can
set these either through an override values file or via the command line when installing the istiod
chart. These are:
values.istio_cni.enabled
should be set to the same value asvalues.cni.enabled
.values.istio_cni.chained
should be set to the same value asvalues.cni.chained
.
For example:
$ helm install istiod istio/istiod -n istio-system --set istio_cni.enabled=true --wait
Hosted Kubernetes settings
The istio-cni
plugin is expected to work with any hosted Kubernetes version using CNI plugins.
The default installation configuration works with most platforms.
Some platforms required special installation settings.
apiVersion: install.istio.io/v1alpha1
kind: IstioOperator
spec:
components:
cni:
enabled: true
namespace: kube-system
values:
cni:
cniBinDir: /home/kubernetes/bin
$ istioctl install --set profile=openshift
Operation details
Upgrade
When upgrading Istio with in-place upgrade, the
CNI component can be upgraded together with the control plane using one IstioOperator
resource.
When upgrading Istio with canary upgrade, because the CNI component runs as a cluster singleton,
it is recommended to operate and upgrade the CNI component separately from the revisioned control plane.
The following IstioOperator
can be used to operate the CNI component independently.
This is not a problem for Helm as the istio-cni is installed separately.
apiVersion: install.istio.io/v1alpha1
kind: IstioOperator
spec:
profile: empty # Do not include other components
components:
cni:
enabled: true
values:
cni:
excludeNamespaces:
- istio-system
- kube-system
When installing revisioned control planes with the CNI component enabled,
values.istio_cni.enabled
needs to be set, so that sidecar injector does not inject the istio-init
init container.
apiVersion: install.istio.io/v1alpha1
kind: IstioOperator
spec:
revision: REVISION_NAME
...
values:
istio_cni:
enabled: true
...
The CNI plugin at version 1.x
is compatible with control plane at version 1.x-1
, 1.x
, and 1.x+1
,
which means CNI and control plane can be upgraded in any order, as long as their version difference is within one minor version.
Race condition & mitigation
The Istio CNI DaemonSet installs the CNI network plugin on every node.
However, a time gap exists between when the DaemonSet pod gets scheduled onto a node, and the CNI plugin is installed and ready to be used.
There is a chance that an application pod starts up during that time gap, and the kubelet
has no knowledge of the Istio CNI plugin.
The result is that the application pod comes up without Istio traffic redirection and bypasses Istio sidecar.
To mitigate the race between an application pod and the Istio CNI DaemonSet,
an istio-validation
init container is added as part of the sidecar injection,
which detects if traffic redirection is set up correctly, and blocks the pod starting up if not.
The CNI DaemonSet will detect and handle any pod stuck in such state; how the pod is handled is dependent on configuration described below.
This mitigation is enabled by default and can be turned off by setting values.cni.repair.enabled
to false.
This repair capability can be further configured with different RBAC permissions to help mitigate the theoretical attack vector detailed in ISTIO-SECURITY-2023-005
. By setting the below fields to true/false as required, you can select the Kubernetes RBAC permissions granted to the Istio CNI.
Configuration | Roles | Behavior on Error | Notes |
---|---|---|---|
values.cni.repair.deletePods | DELETE pods | Pods are deleted, when rescheduled they will have the correct configuration. | Default in 1.20 and older |
values.cni.repair.labelPods | UPDATE pods | Pods are only labeled. User will need to take manual action to resolve. | |
values.cni.repair.repairPods | None | Pods are dynamically reconfigured to have appropriate configuration. When the container restarts, the pod will continue normal execution. | Default in 1.21 and newer |
Traffic redirection parameters
To redirect traffic in the application pod’s network namespace to/from the Istio proxy sidecar, the Istio CNI plugin configures the namespace’s iptables. You can adjust traffic redirection parameters using the same pod annotations as normal, such as ports and IP ranges to be included or excluded from redirection. See resource annotations for available parameters.
Compatibility with application init containers
The Istio CNI plugin may cause networking connectivity problems for any application init containers. When using Istio CNI, kubelet
starts an injected pod with the following steps:
- The Istio CNI plugin sets up traffic redirection to the Istio sidecar proxy within the pod.
- All init containers execute and complete successfully.
- The Istio sidecar proxy starts in the pod along with the pod’s other containers.
Init containers execute before the sidecar proxy starts, which can result in traffic loss during their execution. Avoid this traffic loss with one of the following settings:
- Set the
uid
of the init container to1337
usingrunAsUser
.1337
is theuid
used by the sidecar proxy. Traffic sent by thisuid
is not captured by the Istio’siptables
rule. Application container traffic will still be captured as usual. - Set the
traffic.sidecar.istio.io/excludeOutboundIPRanges
annotation to disable redirecting traffic to any CIDRs the init containers communicate with. - Set the
traffic.sidecar.istio.io/excludeOutboundPorts
annotation to disable redirecting traffic to the specific outbound ports the init containers use.
Compatibility with other CNI plugins
The Istio CNI plugin maintains compatibility with the same set of CNI plugins as the current
istio-init
container which requires the NET_ADMIN
and NET_RAW
capabilities.
The Istio CNI plugin operates as a chained CNI plugin. This means its configuration is added to the existing
CNI plugins configuration as a new configuration list element. See the
CNI specification reference for further details.
When a pod is created or deleted, the container runtime invokes each plugin in the list in order. The Istio
CNI plugin only performs actions to set up the application pod’s traffic redirection to the injected Istio proxy
sidecar (using iptables
in the pod’s network namespace).